Learning to Resolve Conflicts for Multi-Agent Path Finding with Conflict-Based Search

TL;DR

This paper introduces a machine learning-based conflict selection strategy for Conflict-Based Search in multi-agent path finding, significantly improving success rates, search efficiency, and runtime over existing methods.

Contribution

It proposes a novel machine learning framework that learns to select conflicts in CBS, replacing the slow oracle with a fast, effective linear ranking function.

Findings

Improved success rates in multi-agent path finding benchmarks.

Reduced search tree sizes and runtimes.

Outperforms current state-of-the-art CBS solvers.

Abstract

Conflict-Based Search (CBS) is a state-of-the-art algorithm for multi-agent path finding. At the high level, CBS repeatedly detects conflicts and resolves one of them by splitting the current problem into two subproblems. Previous work chooses the conflict to resolve by categorizing the conflict into three classes and always picking a conflict from the highest-priority class. In this work, we propose an oracle for conflict selection that results in smaller search tree sizes than the one used in previous work. However, the computation of the oracle is slow. Thus, we propose a machine-learning framework for conflict selection that observes the decisions made by the oracle and learns a conflict-selection strategy represented by a linear ranking function that imitates the oracle's decisions accurately and quickly. Experiments on benchmark maps indicate that our method significantly improves the success rates, the search tree sizes and runtimes over the current state-of-the-art CBS solver.